Model comparison

MiniCPM-o-4_5 vs Qwen3-30B-A3B-Thinking-2507

At Q4_K_M, MiniCPM-o-4_5 is the smaller download — 5,026,714,400 bytes against 18,556,685,824. At long context the gap widens: Qwen3-30B-A3B-Thinking-2507's KV cache at 32K is 1.5× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

MiniCPM-o-4_5Qwen3-30B-A3B-Thinking-2507
Parameters9.4B30.5B
Architectureqwen3qwen3moe
Layers3648
Native context40,960262,144
Mixture of expertsnoyes, 128 experts
Quantizations published1151
Smallest quantization4.45 GiB7.05 GiB
Q4_K_M4.68 GiB17.28 GiB
Licenceapache-2.0apache-2.0

KV cache by context

the term that decides long-context viability
ContextMiniCPM-o-4_5Qwen3-30B-A3B-Thinking-2507Ratio
4,0960.56 GiB0.38 GiB1.50×
8,1921.13 GiB0.75 GiB1.50×
16,3842.25 GiB1.50 GiB1.50×
32,7684.50 GiB3.00 GiB1.50×
65,5369.00 GiB6.00 GiB1.50×
131,07218.00 GiB12.00 GiB1.50×